cCRMs_seg_classify

The cCRMs_seg_classify project is used for a Machine Learning Pipeline for Automated Segmentation and Classification of Complicated Cystic Renal Masses in MRI.

Installing Dependencies

pip install -r requirements.txt

Data Preprocessing

The input images required for model creation, along with data augmentation operations, are conducted offline.

python [certain modality]_process.py

Segmentation Training

Using the default values for hyper-parameters, the following command can be used to initiate training using PyTorch:

python train_[certain modality].py
--batch_size=2
--nEpochs=500
--classes=2
--inChannels=1
--dropout_rate=0.5
--lr=1e-3
--opt='adamw'
--model='VNET' or 'UNET3D'
--cuda

Segmentation Validation

You can use the following command to initiate model inference using PyTorch:

python eval_[certain modality].py
--batchSz=1
--classes=2
--inChannels=1
--dropout_rate=0.5
--model='VNET' or 'UNET3D'
--resume='model_saved_path/model.pth'
--cuda

Feature Extract

python feature_extract.py

Training Classifier

Using the default values for hyperparameters, you can use the following command to train a random forest.

RandomForestClassifier 超参数搜索空间：
{
  "n_estimators": [2,5,10,15,20],
  "criterion": ["gini", "entropy"],
  "max_features": ["sqrt", "log2", None, 1, 3-10],
  "max_depth": [7-13],
  "min_samples_split": [10-16],
  "class_weight": [{0:0.3, 1:0.75}]
}

Evaluate Classifier

Evaluate the random forest model using the following command.

python eval_RF.py